A transportation or distribution network is a dynamic, stochastic, and complex system. Modelled as a graph, nodes fall into categories that correspond to manufacturing sources, distribution centres, and end customers. In the case of manufacturing enterprises, nodes can represent manufacturing and distribution sites that procure raw material, process them into finished goods, and distribute the finished goods to customers.
The most common objective in transportation network optimization is to find the shortest-distance distribution on a network, i.e., to determine an optimal set of routes between suppliers and customers. Today there is a growing interest in a new and much more sophisticated class of network solutions that involve multiple optimization factors like profit, service level, fault tolerance (or resilience), and environmental footprint, with the optimal solution balancing the complex trade-offs among all these parameters simultaneously.
Supply chains operate in a multi-echelon network that links supply sources, integration and manufacturing points, and end-customers, into a continuous web of commercial activity. At a company of Caterpillar’s scale, this network reaches extraordinarily high levels of complexity, making the problem clearly NP-hard.
Commercial network modelling solutions have existed since the late 1960s, and generally followed a linear solver model. This model has inherent limitations in its ability to 1) scale to problems of our size and 2) accommodate real-world variabilities in critical factors such as transit times, transportation costs, production capacities, and network architecture changes. Caterpillar’s industrial experience, and the experiments run by Caterpillar Logistics Research & Innovation, have shown that ignoring this variability leads to serious errors in the design of global networks.
In our first round of technology development for network optimization, we down selected between multiple technologies to find those most capable of incorporating real world variabilities, tolerating missing data, and accelerating the modelling process over current commercially-available tools. These methods successfully integrated energy price volatility, ocean lane cost commitment terms, cycle stock and in-transit inventory carrying cost implications, and tariff effects. In our next phase of research, we intend to scale up these efforts on additional test problems, which requires additional innovation.
- Development of conceptually new algorithms for transportation network optimization.
- Designing of highly scalable optimization tools.
- Production of consistent optimal results by using proposed algorithms.
- Integration of several dynamic real-world factors, i.e., network resilience, energy volatility models, ocean lane commitment models, etc.
- Frequent outperformance by the computed solutions over the best-known transportation network optimizations produced from commercial tools.
- Development of visualization tools to better present our findings.
See below the presentation by Tony Grichnik on the the outcome and impact of the collaborative research between Industrial AI centre.
Future work on this project will successfully modify the new supply chain optimization processes in order to address four key challenges essential to scaling up these technologies to regular production use.
- Accelerate the prototype supply chain optimization processes while incorporating multiple simultaneous objectives. In particular, we want to temper our profit-based solutions from previous development efforts with a more comprehensive set of resilience considerations to ensure maximal, stable profit over time.
- Incorporate multi-echelon, stochastic inventory carrying costs within the optimization loop. Variabilities in process times can significantly alter the amount of inventory required to meet promised time commitments in a pull-based system. Therefore it is necessary to include safety stock carrying costs with the cycle stock and in-transit inventory carrying costs, without having a significantly adverse effect on solution time.
- Enable variable depth models of global supply chains, leading to “end-to-end” solutions. In our down-selection phase, we tested our algorithms with a traditional four-layer supply chain network (i.e., production sources to outbound ports, to receiving ports, to dealers). Optimization of today’s complex global supply chains requires greater depth, i.e., Tier 1 suppliers to outbound ports, to receiving ports, to production sources, to outbound ports, to receiving ports, to dealers. Other models may have different levels of depth and configuration, so we must modify the processes to become configurable with respect to the depth of the supply chain and transportation network.***
- Manage the coordination of components in a bill of materials, flowing into a distributed production network. At the Tier 1 level in the previous example, production sources require coordinated flows of multiple components. For instance, a test product from our previous round of development has seven (7) key Tier 1 component streams that define a significant percentage of the production cost at those sources. While economies of scale in sourcing consolidation are obvious, it is also true that the more the flow of these Tier 1 components can share the same inbound transportation lanes to the production sources, the lower the transportation cost influence on the overall production cost. This requires an intelligent trade off to be made between costs of source components, costs of transporting those source components to the production sources, and the profitability of the overall network in delivering goods to end-consumers. From a resilience point of view, we must balance the advantages of sharing transportation lanes between the inbound and outbound networks with the risks of simultaneously trapping inbound and outbound flows due to a single point of failure at a production source, port, or transportation lane.
The project team will:
- Modify existing prototype supply chain optimization codes to address the four (4) stated challenges.
- Implement these approaches in prototype form and test them against data sets as provided.
- Collaborate with other researchers inside and outside Caterpillar, who are working on the same challenge.
The Caterpillar Logistics Research & Innovation department will provide:
- Data sets for evaluation.
- Definitions of performance criteria and process constraint measures.
- Access to advanced computing resources if needed and as can be arranged relative to current internal commitments.
- Develop and deliver prototype software and source code implementing the proposed solutions.
- At end of Fall 2013 semester, report out on test results for the first two challenges.
- At end of Spring 2014 semester, report out on project results for the remaining challenges.
- Transportation Network Optimization, Encyclopedia of Business Analytics and Optimization
A. Ogunbanwo, A. Williamson, M. Veluscek, R. Izsak, T. Kalganova, P. Broomhead. Encyclopedia of Business Analytics and Optimization (1st Edition 2014), John Wang, IGI Global. DOI: 10.4018/978-1-4666-5202-6. ISBN: 9781466652026.
- Composite Goal Methods for Transportation Network Optimization
Marco Veluscek, Tatiana Kalganova, Peter Broomhead, Anthony Grichnik. Expert System with Applications, Volume 42, Issue 8, 15 May 2015, Pages 3852–3867. DOI:10.1016/j.eswa.2014.12.017.
- Improving Ant Colony Optimization Performance through Prediction of Best Termination Condition
M. Veluscek, T. Kalganova, P. Broomhead. Industrial Technology (ICIT), 2015 IEEE International Conference, Pages 2394,2402, 17-19 March 2015, DOI: 10.1109/ICIT.2015.7125451
- T. Grichnik, C. Nikolopoulos, A. Byerly, E. Hill, B. Kelly, T. Kalganova, M. Veluscek, P. Broomhead (Filing Date: 15/05/2014) Supply Network Optimization Method and System for Multiple Objectives, CAT Ref.: 08350 1265 (13-1756). Provisional Patent Application.
- T. Kalganova, T. Grichnik, A. Ogunbanwo, A. Williamson, M. Veluscek, R. Izsak, P. Broomhead (Filing Date: 15/03/2014) Supply Chain Network Modeling Using an Evolutionary Strategy, CAT Ref.: 14-0155. Provisional Patent Application.
- T. Kalganova, T. Grichnik, A. Ogunbanwo, A. Williamson, M. Veluscek, R. Izsak, P. Broomhead (Filing Date: 15/10/2013) Hybrid Supply Chain Modelling, Optimisation and Presentation, CAT Ref.: 13-0770. Provisional Patent Application.
- T. Kalganova, T. Grichnik, S. Gaoetswe, and P. Parth (Filing Date: 14/09/2012) Systems and Methods for Forecasting Using Cartesian Genetic Programming, CAT Ref.: 12-1105 and 12-1106; Finnegan Ref.: 08350.1049. Provisional Patent Application.
- T. Grichnik, T. Aguilar, K. Jasti, S. Vamaraju, C. Nikolopoulos, A. Byerly, T. Kalganova, M. Veluscek, and P. Broomhead (Filing Date: 27/03/2015) Method and System for Managing Supply Chain Network with Multiple Supply Layers, CAT Ref. No.: 08350.1735-00000 (14-1941). Provisional Patent Application.